MobiPerf YUANYUAN ZHOU UNIVERSITY OF MICHIGAN MobiPerf Overview - PowerPoint PPT Presentation

MobiPerf YUANYUAN ZHOU UNIVERSITY OF MICHIGAN

MobiPerf Overview Joint work by Google/M ‐ Labs, Dave Choffnes (UW), and Michigan. MobiPerf is a measurement platform for mobile networks , useful to mobile users and researchers Send Measurement Data Run Tests Mobile Users Researchers Return Intuitive Results

Motivation For mobile users:  Have senses of real ‐ time network conditions of the cellular networks  Compare with previous measurement results  Compare measurement results between different carriers/regions  Diagnose network applications

Motivation (Cont.) For researchers:  Measurement results dataset is valuable for researchers working on mobile networks  Open source software (https://github.com/Mobiperf/MobiPerf)  Easily plug ‐ in your customized measurement task  Allow periodic scheduling measurement tasks

Related Work  My Speed Test by Network Operations and Internet Security Lab at Georgia Institute of Technology  Downlink/Uplink throughput test  Latency test  Application bandwidth usage (removed recently)  Tempo by Rice Efficient Computing Group at Rice University  Focus on performance characterization and optimization of browsers on smartphones

MobiPerf System Design End ‐ to ‐ end measurements between end device and central server, mobile networks treated as black boxes MobiPerf MobiPerf Server Suite Client Mobile Networks

MobiPerf System Design (Cont.)  Measurements  Traceroute  HTTP  TCP Throughput (Downlink/Uplink)  Ping  DNS lookup  Server Deployment: MLab

General Challenges about Measurement for Mobile Platforms  User privacy  Accuracy (cellular networks architecture is complicated)  Efficiency (running time, energy consumption, CPU utilization)  Active measurement versus passive measurement

MobiPerf on Android

User Statistics  612 active users  6GB tranmitted measurement data per day  800 requests per day (maximum 3000)

Future Work  Measurement related to IPv6  Compare IPv4 and IPv6 performance  RRC state inference test

Discussions How to further improve MobiPerf with the previously mentioned challenges?  Resource constraints (CPU, energy)  Measurement accuracy  Data analysis support (data uploading, data analyzing)

Thank You!  MobiPerf Website: http://www.mobiperf.com/  To get collected data, you can go to https://storage.cloud.google.com/?arg=openmobiledata_public#openmobiledata_public (all the data); http://openmobiledata.appspot.com/ (your data associated with google id, need to login)